Suggestions about the use AutoDock-GPU with NVIDIA A100 GPUs

[xavgit]

Hi, I have access to a cluster, Leonardo at CINECA, and I have compiled from the sources AutoDock-GPU using OCLGPU as a DEVICE following the issue #239.

At login I receive the following informations: Atos Bull Sequana XH21355 "Da Vinci" Blade -
Red Hat Enterprise Linux 8.6 (Ootpa)

3456 compute nodes with:

• 32 cores Ice Lake at 2.60 GHz
• 4 x NVIDIA Ampere A100 GPUs, 64GB
  • 512 GB RAM

For my needs I'll use few nodes. Can you suggest me some general guidelines about which command line options can be used to get the most out of this kind of GPUs or, in other words, to avoid the underuse of computing power of these cards?

Many thanks.

Saverio

[atillack]

@xavgit Three things:

• compile AD-GPU with OVERLAP=ON to get OpenMP support,
• use a batch file so there's enough work (we typically run with about 10,000 ligands),
• use -D all to have all GPUs work concurrently

[atillack]

one more thing: if you have access to an NVME scratch space it's a good idea to use it to run the dockings node-local on - file io with four A100s will be a bottleneck particularly on a networked file system

[xavgit]

Hi, many thanks for these very useful suggestions.

Saverio